Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

409

International Journal of Agricultural Management and Development Available online on: www.ijamad.iaurasht.ac.irISSN: 2159-5852 (Print)ISSN:2159-5860 (Online)Research Paper

Keywords: FFS extension, train andvisit, tea productivity,demand–driven exten-sion services, propensityscore matching, self-se-lection endogeneity

Received: 26 November 2017,Accepted: 02 August 2019 Agricultural Extension services are among the most im-portant rural services in developing countries. Theservices are considered to be a key driver of technologicalchange and productivity growth in agriculture. In Kenya, likein the rest of the developing economies, agricultural extensionhas largely been delivered through supply–driven approaches.Due to perceived low impact of agricultural extension, thecountry is implementing the National Extension Policy (NEP)which advocates for demand–driven extension and partici-pation of other players. Using the case of the smallholder teasub-sector, this paper examines the effects the FFS extensionon tea crop yields in Kenya. The FFS system uses participatoryapproaches including the demonstration of best sustainablepractices in the farms and farmers learn by doing. Data forthe study was collected from a sample of 525 farm householdsin Western Kenya using a multi stage random sampling pro-cedure and analyzed using the propensity score matching(PSM) model which controls for self-selection endogeneity.The results show that participation in FFS extension increasesannual tea yields by an average of 471.70 kgs per acre(p=0.009) while the farmer–funded train and visit systemhas no influence on crop yields. A part from showing thecontribution of FFS to crop yields, the paper demonstratesthat the supply–driven extension models including T&V arenecessary to stimulate demand in the initial stages of imple-menting the FFS models. Based on the findings, investmentsto enhance FFS access among smallholder farmers are rec-ommended.

Abstract

Does Participation in Farmer Field School ExtensionProgram Improve Crop Yields? Evidence from Small-holder Tea Production Systems in Kenya

Josiah M Ateka a,*, Perez Ayieko Onono - Okelo b and Martin Etyang c

a Jomo Kenyatta University of Agriculture and Technology, Kenyab Kenyatta University, Department of Applied Economics, Kenya c Kenyatta University, Department of Economic Theory, Kenya

Corresponding author’s email: jateka@jkuat.ac.ke

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

410

Does Participation in Farmer Field School.../ Ateka et al.

InTroDuCTIonThe agricultural sector in developing coun-tries is facing unprecedented changes andchallenges, including climate change, landand water constraints and changing marketsand consumption patterns. The knowledgeintensive nature of the sector is thereforemore evident now than ever before. In re-sponse to the need to respond to the emerg-ing challenges, the design of agriculturalextension programs has been the subject ofmuch debate. At the heart of the debate arequestions regarding the choice of extensionmodels that can work ‘best’ for smallholderfarmers (Bitzer et al. 2016; Ektear et al.,2017; Klerkxet al., 2013; Nettleet al., 2017;Swanson & Rajalahti, 2010). A particularissue emerging from scholarly and practi-tioner domains relates to the effectiveness oftwo dominant approaches to extension serv-ices—Training and Visit (T&V) and FarmerField Schools (FFS). The T&V approach relieson the “top-down” extension of technical in-formation, with specialists and field stafftransferring knowledge to “contact farmers”in villages, who in turn are responsible fordiffusing knowledge into the local commu-nity (Davis, 2008; Musa et al. 2013). Themain weaknesses of the system relate to thefact that its supply driven and has beenblamed to be non-responsive to the farmersneeds and interests. Additionally, T&V suffersfrom sustainability issues due to the high costof implementation (Musa et al., 2013). De-spite facing immense criticism, various vari-ants of the T&V system are still beingimplemented in different contexts in the de-veloping world. As a response to the weaknesses of T&Vsystem, FFS was developed as a “bottom-up”approach to extension with a focus on im-proving the problem-solving capacity offarmers through participatory, experiential,and reflective learning (Anderson & Feder,2007). FFS as an approach provides a plat-form for farmers to meet regularly in groupsto study, test and adapt farming practices totheir local conditions (Glendenninget et al.,

2010). The approach differs significantlyfrom mainstream extension practice by itsemphasis on group peer learning, facilitationrather than a teaching pedagogy and local in-novation processes rather than technologicalmessage transfer (Friis-Hansen et al., 2012).Although the FFS approach is being triedand scaled up, its cost-effectiveness and abil-ity to ensure sustained increase in productiv-ity and impact of the approach is still asubject of ongoing debate (Birner et al., 2009;Davis et al., 2012; Larsen & Lilleør, 2014).While some studies find positive impacts ofFFS on agricultural yields, others do not (forexample, Abdullah et al., 2014; Davis et al.,2012; Feder et al., 2004; Friis-Hansen et al.,2012). A major drawback of most previousstudies is that they do not appropriately con-trol for potential differences between FFSparticipants and farmers in the comparisongroup, making it difficult to assign a causal at-tribution to the estimates. The problem isthat participants and non-participants maystill exhibit differences in yields even in theabsence of participation, which renderscausal interpretation of the differences diffi-cult (Verbeek, 2012). Additionally, many pre-vious studies on FFS, do not take into accountthe fact that there are many instances whereFFS programs are implemented alongside theT&V system. The implication is that they aretherefore not able to account for the influ-ence of the T&V on FFS performance. Another weakness is that many previousstudies are based on qualitative methods(Birner & Anderson, 2007; Diab, 2015; Federet al., 2004; Friis-Hansen et al., 2012; Mfitu-mukiza et al., 2017), which despite providingin-depth understanding on how extensionapproach works, fails to provide robust em-pirical evidence on quantitative indicators ofimpact such as yields and incomes. In light ofthe foregoing, many questions about when,where, and how FFSs should be implementedcontinue to trouble extension actors in thedeveloping countries.This article uses a treatment effects frame-work which controls for placement bias and

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

411

Does Participation in Farmer Field School.../ Ateka et al.a unique data set to assess the impact of FFSextension on smallholder tea yields in Kenya.The data was collected in western Kenyawhere the Kenya Tea Development Agency(KTDA), one of the largest smallholderschemes in world (Mbeche & Dorward, 2014)is implementing an integrated extensionmodel that combines both FFS and elementsof the T&V system. The data set allowed us tosimultaneously compare the impacts of FFSand T&V, a contribution that has not beenconsidered in most previous studies. Thepaper shows that participation in FFS is im-portant for tea yields and further demon-strates that while the T&V system appearsnot to have effects on tea yields, it is impor-tant in enhancing participation in FFS. Ourextension to include the determinants of FFSparticipation in the analysis was motivatedby low levels of enrolment in FFS by farmersdespite scale up efforts by KTDA. Drawingfrom the results, the paper reflects on the cir-cumstances under which the two extensionsystems become complimentary rather thansubstitutes. The paper proceeds as follows. Section twodescribes context of FFS implementation inthe tea subsector in Kenya. Section threepresents the theoretical and empirical mod-

els applied in the analysis of impact, whilesection four presents the data and variables.Section five presents the results and discus-sions and section six concludes with reflec-tions on policy implications.FFS implementation in smallholder tea subsec-tor in KenyaTea plays an important role in Kenya’ssocio-economic development. The industry isa leading foreign exchange earner and offerslivelihood to over 0.6 million smallholderfarmers (Mbeche & Dorward, 2014). Despiteits critical role in the economy, productivityhas remained low characterized by stagna-tion and decline (Ateka et al., 2018). Theunimpressive trend reflects existence of pro-duction constraints as evidenced by hugeyield differentials between the smallholderand the plantation tea subsectors (Figure 1).The better performance of the plantationsubsector is often attributed to the presenceof appropriate systems for technology trans-fer including capacity for in-house research.Enhancing yields is an essential factor ofgrowth in the subsector since tea cultivationrequires high investment and involves veryhigh switching costs.

Figure 1. Productivity trends (2003-2012) in Kenya

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

412

Does Participation in Farmer Field School.../ Ateka et al.In response to the productivity downturns,KTDA in the year 2006 and with supportfrom the DFID and other stakeholders startedto pilot the FFS extension approach as a keyintervention. After successful trials and pilot-ing, FFS has been embedded within the KTDAextension strategy and is being implementedto compliment a privatized and farmer–funded T&V system which was introduced inthe tea sector in the late 1990s as part of thestructural adjustment reforms. The priva-tized T&V system incorporated a number ofperformance evaluation elements based onnegotiated key performance indicators thatinclude among others; a specified number offarm visits, farmer meetings, demonstrationsand field days. Despite the initial optimism,the privatized extension system like its pred-ecessor (the public funded T&V system) wascharacterized by challenges in meeting farm-ers’ expectations (Mose et al., 2016). Imple-mentation of the new system thatincorporates FFS with some elements of theT&V is underway (Mose et al., 2016). But de-spite efforts to scaleup, participation in FFSamong the tea farmers has remained low. Keyinformant interviews with KTDA staff indi-cated that FFS enrollment was about 10 per-cent by 2016. The KTDA FFS model focuses on trainingfarmers on sustainable practices through thefield schools which are units where tea farm-ers are trained to practice sustainable agri-cultural practices such as best-practices intea husbandry, soil fertility management, pes-ticide use and protection of bio-diversity. Theapproach empowers farmers to be their owntechnical experts and to adapt potentially ap-plicable technologies to their own particularconditions by enhancing farmers knowledge(technical and socio–economic), decisionmaking and problem-solving skills, and stim-ulating collective action (Feder et al., 2004).Farmers trained in sustainable agriculturalpractices are expected to train others toachieve the required husbandry standards(Friis-Hansen et al., 2012; Glendenning, et al.,2010).

METhoDologYThe theoretical approach and empirical models The analysis of impact in this paper is as-sessed within a treatment effects framework.This refers to the causal effect of a binaryvariable on an outcome variable of policy in-terest (Cameron &Trivedi, 2005) which in thepresent situation is the effect of participatingin FFS on tea yields. To analyze the effectstwo potential outcomes; the outcome withtreatment, and the outcome without treat-ment can be delineated. Using these out-comes, the average treatment effects (ATE)and the average treatment effects on thetreated (ATET) can be derived as shown inequation 2.1 and 2.2.

ATE= E(Y1i - Y0i|Xi) (1) ATET= E(Y1i - Y0i|Xi; Di=1) (2)where, is a vector of household character-istics and denotes the household participa-tion status in an extension program. UnlikeATE which simply describes the expected ef-fect of treatment for an arbitrary householdwith characteristics; ATET measures themean effect of those who actually partici-pated in the program. The measure is there-fore more relevant for evaluating thetreatment effects of a program and isachieved by comparing the performance ofthe participating households with the out-come the same households would haveachieved without participation (Verbeek,2012). A critical issue in the estimation of ATET ishow households are assigned into the treat-ment program. Under random assignment, asis in experimental or quasi experimentalstudies, ATET can be obtained as the differ-ence in the average outcomes between theparticipants and non-participants as shownin equation 2.3;

(3)

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

413

Does Participation in Farmer Field School.../ Ateka et al.The approach is however inappropriate forthe estimation of ATET using cross sectionaldata since the assumption of random assign-ment does not hold. The problem is that theestimates of ATET based on equation 2.3 arebe subject to selection or placement biaseswhich render the estimates causally un-inter-pretable (Cameron & Trivedi, 2005;Clougherty & Duso, 2015). The problemarises when the selection process of theagents being analysed represents an ex-cluded variable that manifests in the errorterm and correlates with the endogenouschoice and outcome variable (Antonakis etal., 2010). In the face of the potential biases,the Propensity Score Matching (PSM) modelwhich provides a non– biased measure of im-pact, under assumptions of conditional meanindependence and common support was ap-plied. The first assumption implies that thedistribution of potential outcomes for theunits under analysis is the same regardless oftheir treatment status. The latter assumesthat all the treated observations have a coun-terpart in the non-treated group (Verbeek,2012).The PSM procedure begins with the estima-tion of propensity scores (PS) which refers tothe probability of treatment given a set of co-variates. The calculated scores are then usedto match the respondents into the two treat-ment groups and was modelled using equa-tion 2.4P(S)i= Pri (Di=1/Xi) (4)where Pri (Di=1/Xi) represents the condi-tional probability of an observation being as-signed into the treatment given its observedcharacteristics. In the study we followed theprevailing trends regarding choice of vari-ables for estimating PS, choice of matching al-gorithms (Khandker et al. 2010), andmatching quality analysis (Caliendo andKopeinig 2008; Rosenbaum & Rubin 1983). After matching, ATET is estimated by calcu-lating the difference in outcome between thetreated and the non – treated group as shown

in equation 2.5. ATET=E(Y1i/D=1) – E(Y0i/ D=1) (5)where E(Y1i/D=1) is conditional mean of theoutcome for the treated contingent on partic-ipation in the treatment and E(Y0i/ D=1) is theconditional mean of the outcome for the non-treated conditional on participation.

Data and variables Data for the study was collected from across sectional survey of 525 smallholder teafarming households drawn from Nyamiraand Bomet counties in western Kenya. Thedata was collected between 2015 and 2016using a multi-stage random sampling proce-dure. The survey collected data on variousfarm level characteristics, household demo-graphic, socio economic characteristics andinstitutional variables. The selection of thevariables was based on previous impact stud-ies and the theory of farm household deci-sion-behavior in developing countries (deJanvry et al., 1991; Fealy & Ahmadpour,2017). This data set is considered uniquesince it was drawn from respondents whohad access to an integrated system of exten-sion that combined both the T&V and FFS ex-tension models. This feature allowedcomparison of the performance of the two ex-tension systems. The definition and measure-ment of variables included in the analysis areoutlined in Table 1.EMPIrICAl rESulTS AnD DISCuSSIonS

Descriptive summaries The results of the descriptive analysis forthe continuous variables are presented inTable 2 while those for the discrete covariatesare summarized in Table3. The descriptiveanalysis includes the comparison of meansbetween the FFS participants and non-partic-ipants for the outcome variable (yields) andthe various covariates of FFS participation.The results (Table 2) show that the partici-pating households had higher tea yields thanthe non-participants. The annual tea produc-

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

414

Does Participation in Farmer Field School.../ Ateka et al.

tivity was 2746 kgs per acre in the combinedsample, compared to 2461 kgs for the non-participating households and 3001 kgs peracre for the participants. While the yield dif-ference provides a hint on the probable direc-tion on the influence of FFS on productivity,the simple comparison of averages has nocausal interpretation, given that the decisionto participate in the FFS program is poten-tially endogenous. The reason is that the dif-ferences in tea yields may not be the result ofparticipation in FFS, but instead might be dueto other factors, such as differences in otherobserved and unobserved characteristics. Acausal attribution would therefore be mis-leading since the participants and non-par-ticipants may still exhibit differences in yieldseven in the absence of participation, render-ing causal interpretation of the difference dif-ficult (Verbeek, 2012).

As shown in Table 2, the farms of the FFSparticipants had higher fertilizer applicationrates; were using less labour per unit area,proportionately more family labour relativeto hired labour and comparatively youngertea bushes than the farms of non-partici-pants. The age of the tea farm is an importantvariable in tea production since aging teaplantations are associated with decline in teaproductivity (Ateka et al., 2018). The averageage of tea bushes in the sample was 27 years.Comparatively, the mean age was 28.5 yearsfor non-participating households and 25.7years for the participants. Additionally, par-ticipating farmers were relatively youngerthan the non-participants. The average age ofthe household head among the non-partici-pants was 50.5 years compared to 48.0 yearsfor the participants. The differences in theother continuous variables including dis-tance from the farm to the nearest market,

Variable Definition and measurement

Extension services (FFS) This is participation of the household in the farmer field school (FFS) program measuredusing dummies; where 1 represents participation in FFS and 0 otherwise.Education The education status of the household head measured in terms of the highest level of edu-cation attained (1= primary, 2 = secondary, 3 = tertiary 4= university).Labour structure The proportion of family labour utilised in tea production. Age of farmer The age of the household head measured in years.Gender This is defined in terms of the gender of the household head and was measured using dum-mies; where 1 represents a male headed household and 0 otherwise.T&V Extension This is the level of extension services received by the farmer measured in terms of the num-ber of visits to/by the extension agent/year.Household assets The value of selected household assets measured in Kenya shillings.Credit Describes receipt of borrowed money for agricultural activities measured using dummies(1= if the household had borrowed and 0 otherwise).Transaction costs Measured using a proxy, the distance to nearest market in kilometres. County dummy The variable accounted for the effects associated with regional differences and was mea-sured using county dummies (1 = Bomet county, 0 = Nyamira county).Age of the tea farm The age of the tea farm measured in number of years since current bushes were planted.Per capita expenditure The per capita amount of money spent on the purchase of a basket of selected householdnecessities per year measured in Kenya shillings.ATMC participation This is participation of the household in an ATMC. The variable was measured using dum-mies; where 1 represents participation in ATMC and 0 otherwise.

Table 1Definition and Measurement of Variables

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

415

Does Participation in Farmer Field School.../ Ateka et al.

per-capita expenditure and household assetswere not statistically significant. The percapita expenditure and household assetswere included in the analysis to represent thehousehold stock of capital which is thoughtto be an important determinant of economicoutcomes and household behavior. It is ex-pected that households with higher capitalstock, would have a higher likelihood to par-ticipate in FFS.Among the discrete variables, the descrip-tive analysis shows that the FFS participatinghouseholds had generally higher levels of ed-ucation than the non-participants. As shownin Table 3, the FFS participants had propor-tionately higher levels of education (sec-ondary, tertiary and university) compared tothe non-participants. In terms of credit use,the results show that the proportion ofhouseholds that had used credit among thenon-participants was 58.1 percent comparedto 79.1 percent for the participants. Thelower levels of credit utilization among thenon-participating households would meanthat they were either more constrained in ac-cessing credit or were more averse to its use.Additionally, the descriptive results show

that proportion of households who had ac-cess to the T&V extension was higher(87.7%) among the participants compared tothe non-participants (75.4%). Determinants of household participation inFFS This section, discusses the results of theprobit model results presented in Table 4.The variables in the model included varioushousehold demographic characteristics (gen-der, age of farmer, education, householdlabour structure), farm level characteristics(farm size, age of farm and distance to thenearest market), the household economiccharacteristic (household assets and percapita expenditure) and institutional vari-ables (access to T&V extension and credit). As shown in Table 4, the coefficient of theage of the household head was statisticallysignificant at standard levels. The coefficientwas negative suggesting that an increase inthe age of the farmer is associated with a de-cline in the probability of participation inFFS. This finding could be attributed to thefact that older farmers are more likely to bereluctant to accept new information and

non -participants(n =248)

Participants(n =277)

Combined(n =525) t p-value

Yield 2461 3001.04 2475.9 3.01** 0.003Fertilizer (bags) 4.28 4.88 4.6 1.78 0.08Labour/acre 175.45 153.52 163.88 1.82 0.07labour structure 51.6 64.03 58.16 3.15** 0.002Age of farm (years) 28.5 25.7 27.0 2.02* 0.032Farm size (hectare) 1.34 1.33 1.34 0.1 0.922Distance (km) 2.86 2.94 2.9 0.33 0.74Household size (percent) 6.47 6.16 6.31 1.24 0.21Age of farmer (years) 50.5 48.0 49.2 1.96 0.05Assets (Kenya Shillings) 112,935.7 102,077.1 107,245.6 1.23 0.22Per capita expenditure 42,548.2 42,752.5 42,658.5 0.073 0.94

Table 2 Summary Statistics for the Continous Variables

**p<0.01 and *p<0.05

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

416

Does Participation in Farmer Field School.../ Ateka et al.

technologies than the younger farmers. In thestudy, a quadratic term, the square of the ageof the household head was introduced to cap-ture the influence of any non-linearities be-tween the relationship between the age of thehousehold head and the participation in FFS.This was necessary since it is expected thatFFS participation would increase to a peak,

Variable Indicatornon-participants Participants Combined

p-value Frequency % Frequency % Frequency %Gender 0 46 18.5 37 13.4 83 15.8 0.104 1 202 81.5 240 86.6 442 84.2Education 1 131 52.8 107 38.6 238 45.3 0.0112 89 35.9 122 44.0 211 40.2 0.0573 20 8.1 30 10.8 50 9.5 0.2814 8 3.2 18 6.5 26 5.0 0.085Credit 0 104 41.9 58 20.9 162 30.9 0.0001 144 58.1 219 79.1 363 69.1T&V extension 0 61 24.6 34 12.3 95 18.1 0.00031 187 75.4 243 87.7 430 81.9Market channel 0 151 60.9 183 66.1 334 63.6 0.2201 97 39.1 94 33.9 191 36.4

Table 3Summary Statistics for the Discrete Variables

Dependent Variable Participation in FFS ExtensionVariable Coefficient Std. Err. Z score p- value

Gender 0.029 0.07 0.43 0.669Age of head -0.023* 0.01 -1.66* 0.096Square of age of head 0.000* 0.00 1.85* 0.064Education (Primary) -0.279** 0.13 -2.15** 0.031Education (secondary) -0.185 0.13 -1.40 0.161Education (Tertiary) -0.109 0.15 -0.73 0.463labour structure 0.168*** 0.06 2.88** 0.004Age of farm (years) 0.018* 0.01 1.68* 0.094Square of age of farm (h) 0.000* 0.00 -1.87 0.062Transaction costs 0.006 0.01 0.68 0.497Household assets 0.000 0.00 -0.75 0.456Per-capita expenditure 0.000 0.00 -0.19 0.846market channels 0.133*** 0.05 2.67*** 0.007T & V Extension 0.214*** 0.06 3.60*** 0.000Credit 0.256*** 0.05 5.06*** 0.000

Table 4Marginal Effects of Determinants of FFS participation

*** p<0.01, **p<0.05 and * p<0.1

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

417

Does Participation in Farmer Field School.../ Ateka et al.as age of the farmer increases then declinesas the farmer gets old or attains retirementage. Although the coefficient of the quadraticterm was positive and statistically significant,the value of its magnitude was zero whichsuggests that the influence of age relatednon-linearities may not have important influ-ence on the household decision to participatein FFS. The finding that the age of the house-hold head is negatively linked with thepropensity of participation in FFS is consis-tent with a study by Asres et al. (2013) whichfound that the age of household head wasnegatively associated the probability of join-ing an extension program.On education, the coefficient of the primaryeducation dummy was negative and statisti-cally significant. The Education level ofhousehold head was included as a proxy forhuman capital and therefore the ability tolearn and apply technologies in tea produc-tion. The results confirm the expectation thatlow levels of human capital (education)would be associated with the likelihood ofnon-participation in the FFS program. Theobservation is consistent with the assertionthat farmers with higher levels of educationare more likely to be the early adopters ofagricultural innovations than those with lesseducation (Gebregziabher et al., 2011). Thefinding is also consistent with Cai et al.(2016) who found that higher educationalachievement and more wealth were associ-ated with attendance of more FFS sessions.The result is also in line with a preference bythe proponents of FFS to target more highlyeducated farmers, those with greater land en-dowments, younger farmers and women andthose with relatively low opportunity costs oflabour (Waddington et al., 2014). It is how-ever important to note that the FFS programcan be used to target low-education groups,since the methodology, demonstration sites,experiential learning methods, group ap-proaches, and other factors can allow peoplewith minimal education to participate andbenefit (Butt et al., 2015).The results of the binary model also show

that an increase in the share of family labourapplied in tea production is associated withan increase in probability of participation inthe FFS program. The finding would be at-tributed to the fact that labour is an impor-tant variable in tea production. Additionally,this would be due to the presence of imper-fections in labour markets which make familyand hired labour to be imperfect substitutes.The reason is that hired labour tends to bemore expensive than family labour due to ad-ditional transaction costs of search andscreenings (Kiani, 2008). Additionally, thehigher transaction costs may be associatedwith the additional monitoring costs sincework effort in a tea farm may not be com-pletely observable, verifiable and enforceable(Ateka et al., 2018). The significance oflabour in tea farming derives from the viewthat when labour constraints are binding,farmers may fail to carry out the required teahusbandry practices at the optimum time orlack the time to attend educational programs. With regard to the farm level characteris-tics, the coefficient for age of the farm wasweakly significant. This means that a farmerwith an older tea farm has a higher probabil-ity of participating in the FFS program than afarmer with younger tea. While an a priori ex-planation for the positive association is lessobvious, the relationship could be related tothe farmer’s experience in tea farming. Thisis because experience broadens the farmer’ssocial network where more market informa-tion can be acquired leading to the establish-ment of more networks and linkages (Shilpi& Umali-Deininger, 2007). The coefficient forthe quadratic term for age of the farm wassignificant but with zero magnitude hence re-futing the importance of non-linearities be-tween age of the farm and participation inFFS. The observation is however inconsistentwith Cai (2016) who found that younger andless experienced farmers were more likely toparticipate in FFS. As shown in Table 4, the coefficients of thevariables included in the analysis to capturethe influence of institutional arrangements

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

418

Does Participation in Farmer Field School.../ Ateka et al.(T&V extension, ATMC participation andcredit) were positive and statistically signifi-cant. The results therefore underscore andconfirm the fact that the presence of appro-priate institutional arrangements includingthe access to input and output markets is akey determinant of farmers’ participation inthe FFS program. This is for instance ex-plained by the observation that borrowingrelaxes liquidity constraints that householdswould face in implementing the new tech-nologies promoted by the FFS curriculum. The other important observation is that theT&V extension has important influence onFFS participation, therefore suggesting exis-tence of complementarities between FFS andother extension systems. This is in light of thelow coverage of extension services in devel-oping countries, which necessitates the needto combine FFS with other faster approachesfor diffusion of information and technologies(Bentley et al., 2015). The implication is thatthe supply–driven extension models includ-ing T&V are necessary to stimulate demandin the initial stages of implementing the FFSmodels. This observation is also important inlight of previous studies showing evidence ofalternatives learning methods having greaterinfluence on the uptake of disseminated tech-nologies than FFS. An example is Ongachi etal. (2017) who found that Video MediatedLearning (VML) was more effective in en-hancing the farmers’ learning and capabilityfor uptake of new agricultural innovationsthan the FFS training. Effects of FFS extension on crop yields After estimation of the PS and the analysisof the determinants of FFS participation,matching of the participant households withthe non-participants was implemented usingthe nearest neighbor matching (NNM) algo-rithm. In NNM, each FFS participant ismatched to closest non-participant using theestimated PS. After the matching process, thequality of the matching was tested using twoprocedures that aimed to assess whether theoverlap or common support condition which

is necessary in PSM was satisfied in the data(Asres et al., 2013). This condition assumesthat some randomness is required in order toguarantee that farm households with identi-cal characteristics can be observed in bothstates (Heckman et al., 1999). The first pro-cedure involved checking the density distri-bution of the propensity scores after thematching process. A visual inspection of thedensity distribution shown in Figure A.1 (inthe appendix) indicates that there was sub-stantial overlap in the distribution of propen-sity scores for both the FFS participating andthe non-participating households. The second procedure involved testing thehypothesis that differences in the means ofthe covariates for both groups were not sig-nificant (after matching). Insignificant differ-ences imply that the matching is successful inbalancing the distribution of relevant vari-ables in both groups. The results of the bal-ancing test are reported in Table 6 in theappendix. The results show there were nosignificant differences in the means of the co-variates for both for the FFS participants andnon-participants after matching. This sug-gests that the specification of the PS wasfairly successful in balancing the distributionof covariates between the two matchedgroups. We therefore conclude that the PSMprocess was suitable for the estimation of theparticipation effects and the common sup-port condition was fulfilled.The effect of FFS participation on TE wascalculated as the difference in average yieldsbetween the two matched and the results arereported as ATET in Table 5.The results in Table 5 show that the coeffi-cient for ATET was 471.7 which suggest thatparticipation in FFS had a positive and statis-tically significant effect on yields. The coeffi-cient for the conventional T&V extension wasnot statistically significant at standard levels.This suggests that there was no statisticallysignificant difference in yields between farm-ers who accessed the T&V extension andthose that that did not. This is despite the factmore households in the sample had accessed

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

419

Does Participation in Farmer Field School.../ Ateka et al.

extension education from T&V system thanthe new FFS extension. The results are con-sistent with expectation since FFS is a newapproach to extension which was adopted bythe tea sector to address the weaknesses ofthe conventional T&V approach (Mose et al.,2016). The attractiveness of FFS is associatedwith its use of participatory adult leaning ap-proaches and emphasis on stronger linkagesbetween research, extension and farmer ex-perimentation. The finding is consistent withevidence showing that that the FFS approachadds to the traditional transfer-of-technologyapproach in imparting knowledge on goodagricultural practices (Ongachi et al., 2017).The finding has equally important implica-tions in light of findings showing that agricul-tural information rather than the farmers’personal and socioeconomic characteristicsis key in maximizing uptake of new agricul-tural interventions (Cai et al., 2016; Ongachiet al., 2017). ConCluSIonS AnD PolICY IMPlICATIonSIn this article, we show the causal effect ofparticipating in FFS-extension on farm yieldsamong smallholder tea farmers in Kenya. Weapplied the PSM model and a variety ofmatching quality analyses which indicatedthat our matching procedures were effectivein balancing the covariates. The study used aunique data set collected to allow for impactevaluation and comparison of two extensionapproaches. One key finding relates to thefarmers’ decision to participate in the FFS

program. The result reveals a fundamentalobservation that while the T&V extensionsystem appears not to have an impact on teayields, its access is important for FFS partici-pation. This may be linked to the ability ofT&V to reach many farmers and its personal-ized attention to specific needs (Jafry et al.,2014). The implication is that for the de-mand-driven extension systems to take rootin practice, farmers must be empowered todevelop their capacity to articulate their de-mands. This is important in light of the lowcoverage of extension services in developingcountries and evidence showing that combin-ing FFS with other methods such video me-diated learning and radio andtelevision-based approaches greatly influ-ences uptake of technologies and innovations(Ongachi et al., 2017; Waddington et al.,2014). The Findings of the study also showthat access to credit and presence of appro-priate arrangements for tea marketing areimportant determinants of FFS participation.We therefore recommend that institutionalactors involved in FFS should consider othercomplementarities that address the farmers’needs in the design and implementation ofthe FFS programme (Waddington et al.,2014). Further, the results show that participationin the FFS program has a positive impact ontea productivity. By providing evidence onthe impact of FFS and the complimentary roleof the T&V, we contribute towards the re-newed discussion regarding the design of ap-

Estimator outcome Effect Coefficient AI robust SE1 Z value p-value

FFS Yield per acre ATET 471.7 181.03 2.61*** 0.009T&V Extension Yield per acre ATET 350.901 240.5872 1.46 0.145*** p<0.011 AI robust standard errors are used to generate heteroskedastic –robust variance estimators to correctfor potential heteroskedasticity (Abadie & Imbens, 2002).

Table 5The Effects of FFS Participation from PSM

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

420

Does Participation in Farmer Field School.../ Ateka et al.propriate extension models for smallholderfarmers in the developing world. Apart fromshowing that FFS is an important pathway forenhancing smallholder crop yields, we alsoshow that scale up and success of the de-mand-driven extension models requiremechanisms that develop and enhance thecapacity of smallholder farmers to articulatetheir extension demands and learning needs(Sulaiman & Blum, 2016). Putting in placesuch mechanisms is therefore an importantprerequisite for the success of FFS. Based on the results, we recommend thedeepening and up scaling of the FFS programso as to enlist more tea farmers into the pro-gram. This should be should be reinforcedwith awareness campaigns to sensitize farm-ers about the enrolment and benefits of theFFS curriculum. We further recommend thatthe design of the FFS program should con-sider complimentary packages includingthose intended to correct for market andother institutional failures.ACKnowlEDgMEnTS This research work was funded jointly bythe Kenya National Research Fund (NRF) andJomo Kenyatta University of Agriculture andTechnology (JKUAT).

rEFErEnCESAbadie, A., & Imbens G. (2002). Simple andbias-corrected matching estimators foraverage treatment effects. Technical Work-ing Paper T0283, NBER.Anderson, J., & Feder, G. (2007). Agriculturalextension handbook of agricultural eco-nomics, agricultural development. Farm-ers, Farm Production and Farm Markets, 3,2343-2378Abdullah, M., Xia, L.C., Li, J., Ghazanfar, S.,Mehmood, Y., Ishaq, M.N., & Saud, S.(2014). Effectiveness comparison betweenthe farmers field school and the training &visit approaches of agricultural extensionin two districts of Pakistan. American-Eurasian Journal of Agricultural & Environ-mental Sciences, 14, 33-39.

Antonakis, J., Bendahan, S., Jacquart, P., &Lalive, R. (2010). On making causal claims:A Review and recommendations. TheLeadership Quarterly, 21(6), 1086–1120.Ateka, J., Onono, P., & Etyang, M. (2018).Technical efficiency and its determinantsin smallholder teaproduction: Evidencefrom nyamira and bomet counties inKenya. Global Journal of Science FrontierResearch: D Agriculture and Veterinary, 18(3), 43–54. Asres, E., Makoto, N., Kumi, Y., & Akira, I.(2013). Effect of agricultural extensionprogram on smallholders’ farm productiv-ity: Evidence from three peasant associa-tions in the highlands of Ethiopia. Journalof Agricultural Science; 5 (8), 163–181.Bentley, J., Chowdhury, A. & David, S. (2015).Videos for Agricultural Extension. Note 6.GFRAS Good Practice Notes for Extensionand Advisory Services. GFRAS: Lindau,SwitzerlandBirner, R. and Anderson, J. R. (2007). How toMake Agricultural Extension Demand-Dri-ven? The Case of India’s Agricultural Ex-tension. Discussion Paper No. 729, IFPRI,Washington, DC.Birner, R., K., Davis, J., Pender, E., Nkonya, P.,Anandajayasekeram, J., Ekboir, A., & CohenM. (2009). From best practice to best fit: Aframework for designing and analyzingpluralistic agricultural advisory servicesworldwide. The Journal of Agricultural Ed-ucation and Extension, 15 (4), 341–355.Bitzer, V., Wennink, B., & de SteenhuijsenPiters, B. (2016). The governance of agri-cultural extension systems. KIT WorkingPaper 2016:1.Butt, T., Gao, Q., & Hussan, M. Z. Y. (2015). Ananalysis of the effectiveness farmer field-school (FFS) approach in sustainable rurallivelihood (SRL): The experience of Pun-jab-Pakistan. Agricultural Sciences, 6,1164-1175.Cai, J., Shi, G., & Hu, R. (2016). An impactanalysis of farmer field school in China.

Sustainability, 8(137), 1-14.Caliendo, M., & Kopeinig, S. (2008). Some

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

421

Does Participation in Farmer Field School.../ Ateka et al.practical guidance for the implementationof propensity score matching. Journal ofEconomic Survey, 22 (1), 31 - 72.Cameron, C., & Trivedi, P. (2005). Micro-econometrics: Methods and applications.Cambridge University Press.Clougherty, J., & Duso, T. (2015). Correctingfor self-selection-based endogeneity inmanagement research: A Review and em-pirical demonstration. Discussion Paper ofDIW Berlin 1465, ISSN Electronic Edition1619-4535.Davis, K. (2008). Extension in Sub-SaharanAfrica: Overview and assessment of pastand current models, and future prospects,15 (3), 15-28.Davis, K., Nkonya, E., Kato, E., Ayalew, D.,Odendo, M., Miiro, R., & Nkuba, J. (2012).Impact of farmer field schools on agricul-tural productivity and poverty in EastAfrica. World Development, 40(2), 402-413.de Janvry, A., Fafchamps, M., & Sadoulet, E.(1991). Peasant household behavior withmissing markets: Some paradoxes ex-plained. Economic Journal, 101(409),1400-17.Diab, A. (2015). Learning impact of farmerfield schools of integrated crop–livestocksystems In Sinai Peninsula, Egypt. Annalsof Agricultural Sciences, 60(2), 289–296.Ektear, U. M., Kanak, P., Gao, A., Habibur, Q.,Rahman, M., & Shafiqul, M. (2017). Effectsof community-based paid extension on re-ducing vulnerability of smallholder dairyfarmers of Southwest Bangladesh. Interna-tional Journal of Agricultural Managementand Development, 7(3), 293-304.Friis-Hansen, E., Duveskog, D., & Taylor, E.(2012). Less noise in the household: Theimpact of farmer field schools on genderrelations. Journal of Research in Peace, Gen-der and Development, 2 (2), 044–055Fealy, N., & Ahmadpour, A. (2017). Effectivefactors on the participation of MazandaranProvince rural women in rice cultivation.International Journal of Agricultural Man-agement and Development, 7(3), 275-282.

Feder, G., Murgai, R., & Quizon, B. (2004).Sending farmers back to school: The im-pact of farmer field schools in Indonesia.Review of Agricultural Economics, 26 (1),45–62.Gebregziabher, K., Mathijs, E., Maertens, M.,Deckers, J., & Bauer, H. (2011). Extension-participation, household income and in-come diversification: A system equationsapproach. Retrieved fromwww.csae.ox.ac.uk/conferences/2011-EDiA/.../437Glendenning, C. J., Babu, S., & Asenso-Okyere,K. (2010). Review of agricultural extensionin India: Are farmers’ information needsbeing met? (No. 1048). International FoodPolicy Research Institute (IFPRI).Heckman, J. J., LaLonde, R. J., & Smith, J. A.(1999). The economics and econometricsof active labor market programs. In Hand-book of Labor Economics (Vol. 3, pp. 1865-2097). Elsevier.Jafry, T., Kingiri, A., & Nderitu S. (2014). As-sessment of extension and advisory meth-ods and approaches to reach rural women– examples from Kenya. Discussion Paper,United States Agency for International De-velopment (USAID) project “ModernizingExtension and Advisory Services (MEAS).www.meas-extension.orgKhandker, S., Koolwal, G. & Samad H (2010).Handbook on Impact Evaluation: Quanti-tative Methods and Practices. Washington:World Bank Publications.Klerkx, K., Koutsouris, A., & Labarthe, P.(2013). Guest. The Journal of AgriculturalEducation and Extension, 19(3), 217-221.Kiani, A. K. (2008). Farm size and productiv-ity in Pakistan. European Journal of SocialSciences, 7 (2), 42-52.Larsen, A., & Lilleør A. (2014). Beyond thefield: The impact of farmer field schools onfood security and poverty alleviation.World Development, 64, 843-859Mbeche, R., & Dorward, P. (2014). Privatiza-tion, empowerment and accountability:What are the policy implications for estab-lishing effective farmer organizations?

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

422

Does Participation in Farmer Field School.../ Ateka et al.

Land Use Policy, 36, 285-295.Mose, G. N., Mbeche, R., & Ateka, J. (2016). In-stitutional innovations for smallholderagricultural production systems in Kenya:A Case of smallholder tea subsector. Euro-pean Journal of Sustainable Development 5,(3), 461-475.Musa, N, Aboki, E., & Audu, I. (2013). The lim-itations and implications of Training andVisit (T&V) extension system in Nigeria.Journal of Agriculture and Sustainability,4(1), 67-76.Mfitumukiza, D., Barasa B., Nankya, A., Nab-wire, D., Owasa, H., Babu, S., & Kato, G.(2017). Assessing the farmer field school’sdiffusion of knowledge and adaptation toclimate change by smallholder farmers inKiboga District, Uganda. Journal of Agricul-tural Extension and Rural Development,9(5), 74-83.Nettle, R., Klerkx, L., Faure, G., & Koutsouris,A. (2017). Governance dynamics and thequest for coordination in pluralistic agri-cultural advisory systems. The Journal ofAgricultural Education and Extension, 23(3), 189-195. Ongachi, W., Onwonga, R., Nyanganga, H., &Okry, F. (2017). Comparative analysis ofvideo mediated learning and farmer fieldschool approach on adoption of striga con-trol technologies in Western Kenya. Inter-national Journal Agricultural Extension,5(1),1-10.Rosenbaum, P. R., & Rubin, D. B. (1983). Thecentral role of the propensity score in ob-servational studies for causal effects. Bio-metrika, 70(1), 41-55.Shilpi, F., & Umali-Deininger, D. (2007).Where to sell? Market facilities and agri-cultural marketing. Policy Research Work-ing Pape, 4455. Washington DC: WorldBank.Sulaiman, R. V., & Blum, M. L. (2016). Tailor-ing rural advisory services for family farms.FAO Policy Paper, Global Forum for RuralAdvisory Services, Rome.Swanson, B. & Rajalahti, R. (2010). Strength-ening agricultural extension and advisory

systems, procedures for assessing, trans-forming, and evaluating extension sys-tems. International Bank forReconstruction and Development, Wash-ington D.C. USA.Swanson, B. (2008). Global review of goodagricultural extension and advisory serv-ice practices. Food and Agriculture Organ-ization of the United Nations (FAO), Rome.Verbeek, M. (2012). A Guide to modern Econo-metrics. 4th Edition, Wiley, New York.Waddington, H., Birte, S., Jorge, H., Martina, V.,Daniel, P., Philip, D., & Howard, W. (2014).Farmer field schools for improving farm-ing practices and farmer outcomes: A sys-tematic review. Campbell SystematicReviews, 6, 1–335.

Inte

rnat

iona

l Jou

rnal

of A

gric

ultu

ral M

anag

emen

t and

Dev

elop

men

t, 9(4), 4

09-423, Decem

ber 2019.

423

Does Participation in Farmer Field School.../ Ateka et al.

APPEnDIx

Variable Mean t-value p-valueparticipants non participants Gender 0.86 0.79 2.09 0.37Age of head 48.24 48.55 -0.25 0.805Education (Primary) 0.4 0.44 0.89 0.38Education (secondary) 0.45 0.42 0.62 0.54Education (Tertiary) 0.1 0.07 1.26 0.21Education (Education) 0.05 0.07 0.89 0.38labour structure 0.68 0.63 0.7 0.48Age of farm 21.03 21.56 0.44 0.66farm size 1.27 1.33 0.64 0.522Transaction costs 2.96 3.13 0.7 0.48Household assets 100,000 110,000 0.67 0.506Per-capita expenditure 40433 37275 1.32 0.188market channels 0.34 0.28 1.61 0.107T & V Extension 0.88 0.89 0.41 0.68Credit 0.78 0.77 0.32 0.75

Table 6Balancing Test of Matched Groups (PSM)

Figure A.1: Distribution of propensity scores for the participants and non-partici-pants after matching how to cite this article:Ateka, J.M., Onono- Okelo, P.A., & Etyang, M. (2019). Does participation in farmer field schoolextension program improve crop yields? Evidence from smallholder tea production systemsin Kenya. International Journal of Agricultural Management and Development, 9(4), 409-423.url: http://ijamad.iaurasht.ac.ir/article_670334_9a9d01d89539d32408a539312cee8aa4.pdf